Hypoglycemia-Associated Changes in the Electroencephalogram in Patients With Type 1 Diabetes and Normal Hypoglycemia Awareness or Unawareness

In type 1 diabetes, the major limiting factor in achieving glucose targets is risk of severe hypoglycemia (1). Impaired hypoglycemia awareness (reduced ability to perceive the onset of hypoglycemia) is associated with a 6- to 20-fold increased risk of severe hypoglycemia (2,3). It is assumed that episodes of repeated mild symptomatic and asymptomatic hypoglycemia contribute to the development of impaired hypoglycemia awareness (4,5). The condition is characterized by loss of hypoglycemia warning symptoms and blunted counterregulatory hormone responses to low blood glucose (6–8). Thus, the threshold at which the patients experience symptoms of hypoglycemia will gradually decrease. For some individuals, this level is equal to or below the threshold for neuroglycopenia (9,10).

The hypoglycemia-associated neuroglycopenia, resulting in neuroglycopenic symptoms and cognitive dysfunction, is mirrored in the electroencephalogram (EEG) by an increased activity in the low-frequency bands (θ and delta band) (11). These changes disappear when the glucose concentration is slightly increased, even before normoglycemia is restored (12,13). This is in contrast to hypoglycemia-induced cognitive dysfunction, which is present up to 75 min after glucose levels are restored (14).

While differences in symptomatic and hormonal counterregulatory responses between subjects with normal or impaired awareness are well known, less is known about differences in neuroglycopenia as determined by EEG and cognitive evaluation during hypoglycemia and recovery following hypoglycemia. The aim of our study was to assess whether hypoglycemia awareness status is associated with differences in the hypoglycemia-induced EEG changes during and shortly after an episode of mild, insulin-induced hypoglycemia in hypoglycemia-aware and -unaware patients with type 1 diabetes.

The study was a clinical controlled study. The protocol is registered at http://clinicaltrials.gov (NCT01337362) and was approved by the Regional Committee on Health Research Ethics. Written informed consent was obtained from all participants.

Twenty-four patients with type 1 diabetes were recruited from the diabetes outpatient clinics at Nordsjællands Hospital Hillerød and Steno Diabetes Center, Denmark. Inclusion criteria were type 1 diabetes for >5 years, age >18 years, and being either hypoglycemia aware or unaware. Exclusion criteria included pregnancy; breastfeeding; any brain disorder; use of antiepileptic drugs, β-blocking drugs, or neuroleptic drugs; use of benzodiazepines within the last month; cardiovascular disease; and alcohol or drug abuse. Hypoglycemia awareness status was classified by the Pedersen-Bjergaard method (15), the Gold score (16), and the Clarke method (17). Of the 24 participants, 14 patients were classified as hypoglycemia unaware and 10 patients as hypoglycemia aware according to all three methods (Supplementary Table 1). Patients who did not qualify as either hypoglycemia aware or unaware were excluded from participation.

From 5 days before the hypoglycemic clamp, the participants wore a continuous glucose monitor (CGM) (Guardian Real-Time with Enlite sensor; Medtronic, Minneapolis, MN) to detect any hypoglycemia in the days before the experiment. In order to reduce the risk of hypoglycemia, the CGM was set to alarm the participant if glucose levels fell <4.5 mmol/L. This was done to compensate for possible inaccuracies and to allow the patient time to take corrective measures before the glucose concentration fell <3.5 mmol/L. All patients used the Contour Link blood glucose meter (Bayer HealthCare, Leverkusen, Germany) for blood glucose monitoring and calibration of the CGM.

On the day of the experiment, subjects arrived in the clinical research unit after an overnight fast. In case of glucose measurements <3.5 mmol/L in the preceding 24 h, study procedures were postponed 2 weeks.

The glycemic targets were 5–6 mmol/L during normoglycemia, 2.0–2.5 mmol/L during hypoglycemia (nadir 2.2), and 5–6 mmol/L during recovery (Fig. 1). For the clamp procedure, insulin (Actrapid; Novo Nordisk, Ballerup, Denmark) mixed with heparinized plasma from the patient and isotonic saline was administered intravenously at a rate of 1 mU insulin/kg/min. A variable 20% glucose infusion was administered to keep plasma glucose at the desired levels.

Digital EEG was measured continuously (Cadwell, Kennewick, WA). The electrodes were placed according to the 10–20 system using electrocaps. Data were collected at a sampling rate of 200 Hz and filtered using a first-order high-pass filter of 0.5 Hz and a first-order low-pass filter at 70 Hz. Explicit care was taken to obtain 5 min of electroencephalographic standard conditions with eyes closed during two specific time points at each glycemic level subsequently analyzed in tandem. Results from P3-C3 electrodes are reported. This location in the parieto-central brain region was chosen because hypoglycemia-associated EEG changes are most abundant in this area (18).

Analyses of the EEG were performed by quantitative EEG (qEEG) analysis focusing on frequency characteristics of the data in the 1) θ band (4–7.75 Hz), where activity is associated with drowsiness, mediation, or light sleeping; 2) α band (8–12.75 Hz), where activity is associated with relaxation with eyes closed; and 3) a combined α-θ band (4–12.75 Hz). The α-θ band was included in order to detect whether any shift in frequency occurred in the transition between the α and θ band that might not otherwise be identified. Power spectral density of the normoglycemic, hypoglycemic, and recovery periods was estimated using the Welch method applied to 4-s Hamming-windowed epochs with 50% overlap and a 0.25-Hz resolution. From these, average amplitude spectra were calculated (from the square root of the power), and absolute amplitude and the centroid frequency were calculated for each band. The absolute amplitude was determined using a numerical integration technique. The centroid frequency was defined as the center of gravity of each frequency band that subdivides the area into two of equal size. Signal processing was performed in Matlab 7.12.0 (MathWorks, Natick, MA). Subsequently, artifact-free analyses were performed on a subset of the standardized EEG recordings. For this purpose, the recordings were visually inspected in a blinded fashion, and 10 segments of a minimum of 4 s without artifacts were identified from the 5-min-long recording. qEEG analysis was performed on the identified segments, and then the mean for each glycemic level was determined.

Three different tests were used at each glycemic period: 1) the Danish version of the Mini CalCAP test (E.N. Miller, California Cognitive Assessment Package, Norland Software, Los Angeles, CA, 1990), which consists of three different reaction-time tasks with increasing complexity in which the subject is asked to identify the number 7 (RT1), two identical numbers (RT2), and two increasing numbers (RT3) in a sequence; 2) Trail 5 B of the Comprehensive Trail Making Test (TMT B) (Proed, Austin, TX), which tests how fast the participant can connect numbers and letters in alternating increasing sequence (i.e., 1-A-2-B, etc.); and 3) the Danish version of the Stroop Color and Word Test by Golden (19) in which the participant must name as many items as possible in 45 s at three different conditions: color names printed in black, blocks printed in different colors, and color names printed in nonmatching colors (20).

The CalCAP tests assess cognitive domains such as recognition memory, attention (focused, divided, and sustained), and processing speed. The TMT B evaluates visual attention, motor speed, and cognitive alternation, and the Stroop tests evaluate processing speed and selective attention.

Throughout the clamp, plasma glucose was analyzed using YSI 2300 (YSI Inc./Xylem Inc., Yellow Springs, OH). At the end of each glycemic period and in the beginning of the hypoglycemic period, blood was drawn for measurement of glucagon, epinephrine, norepinephrine, cortisol, and growth hormone. For analysis of plasma glucagon concentration, samples were extracted in 70% ethanol before measurement, and glucagon was measured with a radioimmunoassay directed against the COOH terminus of the glucagon molecule (antibody code number 4305). Serum growth hormone concentrations were measured by a two-sided immunometric sandwich method with chemiluminescence (Immulite 2000; Siemens AG, Munich, Germany). Blood for measurement of catecholamines was drawn in tubes coated with EGTA, glutathione, and NaOH and measured by an ELISA kit (Labor Diagnostika Nord GmbH & Co. KG, Nordhord, Germany). All samples were centrifuged and stored at −80°C except for cortisol, which was analyzed immediately after the experiment using immunochemistry (Siemens Advia Centaur XP, range 5.5–2,069 nmol/L; Siemens AG, Munich, Germany).

Patients filled out a standardized hypoglycemia symptom questionnaire (a Danish modification of the Edinburgh Hypoglycemia Scale) (21) twice during normoglycemia, hypoglycemia, and recovery. The symptoms were subsequently grouped into three symptom categories: autonomous, neuroglycopenic, and other.

The within-group effects of hypoglycemia were assessed by repeated-measures ANOVA. If the assumption of sphericity was violated, assessed by Mauchly test, a multivariate ANOVA or a Wilcoxon signed rank test was performed as appropriate. The between-group effects were assessed by Student t tests, mixed-model ANOVA test, or a Mann-Whitney U test depending on distribution, occurrence of repeated measurements, and homogeneity of intercorrelations. For the above-mentioned tests, the absolute amplitude measures in the EEG variables were transformed to the logarithmic scale. The data were analyzed using IBM SPSS Statistics, version 20 (IBM Corporation, Armonk, NY), while the mixed-model ANOVA was analyzed using R 2.15.1 (R Foundation for Statistical Computing, Vienna, Austria). A P value <0.05 (two-sided) was considered statistically significant.

The EEG recording failed in one patient (aware), and he was therefore excluded from this study. The baseline characteristics of the remaining 23 participants are shown in Table 1. There was a trend toward the unaware group being older, and they had had diabetes for a longer duration of time. The unaware group also required a lower daily insulin dose, had a lower BMI, and had experienced more episodes of severe hypoglycemia during their lifetime, while glycemic control as measured by HbA_1c did not differ between groups.

During hypoglycemia, plasma glucose was 2.5 (0.05) mmol/L with a nadir of 2.3 (0.07) mmol/L in the aware group and 2.3 (0.02) mmol/L with a nadir of 2.0 (0.04) mmol/L in the unaware group (Fig. 2). The differences in glucose levels between the groups during hypoglycemia were significant (P = 0.04; nadir: P < 0.001) despite a lower glucose infusion rate (GIR) during hypoglycemia in the aware group (P = 0.02). During recovery, the plasma glucose levels were similar, 5.6 (0.09) mmol/L and 5.6 (0.10) mmol/L in the aware and unaware groups (P = NS), but the mean GIR was still higher in the unaware group (P = 0.03).

During hypoglycemia, the absolute amplitude in the θ band up to doubled (aware: normoglycemic, 49 [15] µV [mean (SEM)]; hypoglycemic, 83 [25] µV, P = 0.008; unaware: normoglycemic, 45 [14] µV; hypoglycemic, 112 (39) µV, P = 0.003). Likewise, the log(absolute amplitudes) of the θ band (Fig. 3) and of the α-θ band were increased and the centroid frequency in the α-θ band was decreased in both groups during hypoglycemia (Table 2 and Supplementary Table 2). During recovery, the absolute amplitude in the θ band remained increased in both groups, as did the centroid frequency in the α-θ band in the unaware group (Table 2 and Supplementary Table 2). Analyses of differences between the aware and unaware groups did not reveal differences for any of the examined variables (all P = NS).

The artifact-free analyses (Supplementary Table 3) showed results similar to those for the crude analysis. It identified an increase in the absolute amplitude in the θ band and decrease in the centroid frequency in the α-θ band in both groups during hypoglycemia (all P < 0.05). The unaware group also had an increase in the absolute amplitude in α and α-θ band (α, P = 0.02; α-θ, P < 0.001) and a decrease in centroid frequency in the α band (P = 0.02). There were no differences between the two groups for any of the variables. In the recovery period, the centroid frequency in the α-θ band was still decreased in the aware group (P = 0.03), whereas in the unaware group, the absolute amplitude in the θ band remained higher (P = 0.03).

In order to see whether differences in patient characteristics between the two groups had any effect on the results of the qEEG analysis, a mixed-model ANOVA was performed. We found an increase in absolute amplitude in the θ band during both hypoglycemia (P < 0.001) and recovery (P = 0.001) and in the α-θ band during hypoglycemia (P < 0.001). When adjusting for awareness status, BMI, and either age or diabetes duration, the differences remained (all P ≤ 0.001), while none of the covariates were significantly associated with any of changes in amplitudes. The centroid frequency in the α-θ band also decreased during hypoglycemia (P < 0.001) and recovery (P = 0.02) independent of awareness status, BMI, or age. The other EEG variables did not change during hypoglycemia and were not associated with awareness status, BMI, age, or duration of diabetes.

The aware group had an increase in all three categories of symptom scores during hypoglycemia (autonomous, P = 0.001; neuroglycopenic, P = 0.006; others, P = 0.02) (Fig. 4). In the unaware group, there were also small increases in autonomous and neuroglycopenic symptom scores during hypoglycemia (autonomous, P = 0.02; neuroglycopenic, P = 0.005). When comparing the two groups, the aware group scored higher in all three categories (all P < 0.05). During recovery, the scores had normalized for neuroglycopenic and other symptoms in both groups. In contrast, autonomous symptoms scores were still increased in both groups in the beginning of the recovery period (aware, P = 0.03; unaware, P = 0.04).

In the CalCAP test, the first and most simple task (RT1) showed ∼5% prolonged median reaction time in both groups and a trend toward higher error rates in the unaware group during hypoglycemia. For RT2 only, the unaware group had prolonged reaction time (4%) and a higher error rate. For RT3, there was a trend toward a prolonged reaction time in the aware group (6%) with no difference in error rate in any of the groups. When comparing the two groups’ response to RT1–3, there was no significant difference between the two groups. In the Stroop and TMT B tests, performances deteriorated up to 30% during hypoglycemia with no difference between the groups (Table 3).

During the recovery period, the performances in the Stroop and TMT B tests showed no difference from the normoglycemic level in both groups. In contrast, the CalCAP test showed a longer reaction time during recovery compared with the normoglycemic period in RT1 in both groups (aware, 17%; unaware, 6%) and RT3 in the aware group (8%). The aware group also had a trend toward a higher error rate in RT1. There was no difference in error rate for RT2 and RT3 between normoglycemia and recovery or between the two groups.

Counterregulatory hormones increased in both groups during hypoglycemia (Fig. 5). The increases in epinephrine and cortisol levels from normoglycemia to hypoglycemia were higher in the aware group (epinephrine, P = 0.01; cortisol, P = 0.005). During recovery, all levels normalized apart from cortisol and norepinephrine levels, which remained elevated in the aware group (cortisol, P = 0.04; norepinephrine, P = 0.02).

Our study demonstrates dissociation between EEG changes and symptoms of hypoglycemia as well as hormonal counterregulation in hypoglycemia-aware and -unaware patients with type 1 diabetes during insulin clamp–induced mild hypoglycemia. Thus, awareness status does not affect the hypoglycemia-associated EEG changes despite the fact that the aware group reported higher scores for autonomous and neuroglycopenic symptoms and had higher epinephrine and cortisol responses compared with the unaware group. Moreover, cognitive function deteriorated as expected during hypoglycemia but, as for the EEG changes, no differences were detected between the two groups. This suggests that both hypoglycemia-associated EEG changes and cognitive function are dissociated from the perception of hypoglycemia symptoms and the hormonal counterregulatory responses.

The EEG analyses showed that changes during hypoglycemia were most significant for the absolute amplitude in the θ band. This is in accordance with other studies investigating hypoglycemia-associated EEG changes (11, 22, 23). Increased amplitude in the θ band is not unique to hypoglycemia but is seen during a number of conditions including drowsiness, encephalopathy, or depression (24–26). However, in contrast, hypoglycemia-associated EEG changes are most pronounced in the parietotemporal area of the brain, whereas the perspective of the full EEG is needed in the other conditions.

Our finding that hypoglycemia-associated EEG changes are not effected by awareness status implies that the cortex and the brain as a whole does not adapt to hypoglycemia in the same manner as the sympathoadrenal system and the deeper areas of the brain (27). This finding supports preliminary data (28) showing no difference in hypoglycemia-associated EEG changes between aware and unaware patients and is in accordance with findings from Tribl et al. (18) reporting similar EEG changes in patients with “good” and “poor” awareness at glucose levels <2.8 mmol/L. It is however possible that the hypoglycemia-associated EEG changes are modulated by mild or asymptomatic hypoglycemia. In line with this both Tallroth et al. (13) and Amiel et al. (29) reported that the hypoglycemia-associated EEG changes were weaker or not present in nondiabetic control subjects in contrast to in participants with diabetes or insulinoma.

The study by Tribl et al. (18) did however report differences between their two groups at glucose levels between 2.8 and 3.3 mmol/L and baseline (3.8–7.6 mmol/L). In contrast, we did not find any difference in the EEG between aware and unaware during normoglycemia. Since we studied participants at fixed glucose levels during normoglycemia and hypoglycemia, we might have missed subtle differences that would have been apparent at higher glucose levels like those shown by Tribl et al. (18). However, the onset of hypoglycemia-associated EEG changes show great variation between subjects, varying from 1.6 to 3.4 mmol/L (29,30), and we therefore chose a glucose level during hypoglycemia at which we would expect EEG changes to be present.

The EEG variables did not revert to baseline levels during the recovery period even though glucose levels were restored. This finding contradicts previous EEG studies by Bryan et al. (31) showing that the EEG in rats normalizes during recovery and by Tallroth et al. (13) describing that the EEG in humans normalizes immediately after an episode of insulin-induced hypoglycemia. We find that the amplitude in the θ band is still increased in the recovery period in both groups and that the unaware patients also have increased activity in the α-θ band, suggesting that the brain needs time to recuperate and that the metabolism is altered for at least 1 h after an episode with hypoglycemia. This is in line with studies showing increased glycogen availability in the rodent brain following episodes of hypoglycemia, leading to the theory of brain glycogen supercompensation (32,33). In our study, however, unaware patients had delayed normalization of centroid frequency of the α-θ band, which may support that unaware patients need a longer recovery time. An alternative, more straightforward explanation might be that the unaware group unintentionally was slightly but significantly more hypoglycemic during the clamp. Cognitive function tests showed that reaction time was prolonged during recovery in two of the CalCAP tests in the aware group and one test in the unaware group. The worsening in performance during recovery could be explained by difficulty concentrating due to a carryover effect from hypoglycemia or an adverse response to the test.

Recurrent episodes with hypoglycemia lead to hypoglycemia-associated autonomic failure with blunted hormonal counterregulatory response and hypoglycemia unawareness (34), which may be considered both an adaptive response and a maladaptive response (35). On one hand, the blunted responses may increase the risk of severe hypoglycemia (3). On the other hand, recurrent hypoglycemia may cause the brain to adapt to enduring hypoglycemia (36). Both animal and human studies support the latter theory (37–39). In contrast to these functional neuroimaging studies and animal studies, we did not find any hypoglycemia-associated EEG differences between aware and unaware patients. However, functional neuroimaging techniques and animal studies better target deeper layers and specified areas of the brain than the EEG, which records electrical activity mostly originating from the cortex. A dissociation between the cortex as a whole and the deeper or more specific areas of the brain may thus exist when responding to hypoglycemia unawareness.

The results from this study may have clinical significance. Our observation that hypoglycemia-associated EEG changes are independent of hypoglycemia awareness status indicates that EEG monitoring, as previously suggested, potentially constitutes a method for a hypoglycemia alarm that will also benefit the unaware patients carrying the greatest risk of severe hypoglycemic events (30).

The strengths of our study are that cognitive function, EEG, symptom scores, and hormonal counterregulatory responses were recorded, enabling a broader perspective on neuroglycopenia in hypoglycemia aware and unaware patients. Moreover, classification of awareness status was based upon three validated methods (40), resulting in two clearly separated groups as subsequently confirmed by the differences in hypoglycemia symptom scores during hypoglycemia as well as in the counterregulatory responses. CGM was also applied in order to avoid any symptomatic and asymptomatic hypoglycemia 5 days prior to the study, which could potentially blunt the symptomatic and hormonal responses to hypoglycemia induced during the study (34). Finally, EEG was recorded during standardized conditions and analyzed both on the complete 5-min recording and on artifact-free segments to exclude the possibility that any changes during hypoglycemia, identified in the analyses, were due to artifacts such as movement. The small differences between the results of the two analyses were to be expected. The total 5-min EEG is more robust due to the length of the recording, but presence of artifacts cannot be excluded. In contrast, the quality of artifact-free EEG recording is optimized, but the lengths of the segments are relative short, and brief changes may upset the analysis. Both analyses did, however, identify similar changes in the θ and α-θ band consistent with hypoglycemia-associated EEG changes.

There are also limitations to the study. Since the target glucose was similar for all participants, precise thresholds for onset of hypoglycemia-associated EEG changes and cognitive dysfunction could not be determined, and possible differences between the groups could therefore not be detected. Thus, further studies are needed to address whether hypoglycemia-associated EEG changes can be recorded at higher plasma glucose levels in aware patients compared with unaware. Furthermore, the aware group was examined at a slightly higher glucose level during hypoglycemia compared with the unaware group even though the GIR during hypoglycemia was lower in the aware group. The difficulty in reaching the same nadir glucose in both groups, even during a glucose clamp, is partly a consequence of the better capacity of aware patients to counterregulate. A control day with glucose levels within the normal range was deliberately not performed. Previous studies at similar glucose clamp settings found that EEG changes and cognitive dysfunction were only observed at hypoglycemia while these parameters did not differ from baseline throughout the euglycemic control day (13,41). A post hoc analysis of the data from Høi-Hansen et al. (41), applying the same methods for qEEG analysis as used in this study, also showed that hypoglycemia-associated EEG changes could only be identified on the hypoglycemic study day and not on the euglycemic control day. The EEG changes reported in our study can therefore be attributed to the effect of hypoglycemia on the brain. There were differences between the two groups at baseline. The aware group was younger with shorter duration of diabetes, which was to be expected since both age and duration of disease are well-known risk markers for hypoglycemia unawareness (2,42). However, the aware group also had long duration of diabetes, and the mixed-model ANOVA showed that age and duration of diabetes do not affect hypoglycemia-associated EEG changes, which is also supported by the literature (28). The difference in age could, however, affect the results of the cognitive tests (43) and mask if the unaware group performed better in the TMT B or Stroop Word/Color test.

In patients with long-standing type 1 diabetes, hypoglycemia-induced EEG changes and cognitive performance during hypoglycemia are not affected by hypoglycemia awareness status during a single insulin-induced episode with hypoglycemia. This finding contrasts the expected lower hypoglycemia symptom scores and blunted epinephrine and cortisol responses in the unaware group. The EEG did not normalize during 1 h of recovery even though glucose levels returned to normal, which suggests that the brain needs time to recuperate after an episode with hypoglycemia.

Acknowledgments. The authors thank the participants for being a part of the study and research nurses Pernille Banck-Petersen, Tove Larsen, and Charlotte Hansen, Department of Cardiology, Nephrology and Endocrinology, Nordsjællands Hospital Hillerød, for skillful technical assistance.

Funding. This study was partially funded by research grants from the University of Southern Denmark, the Danish PhD School of Endocrinology, the Research Foundation at Nordsjællands Hospital, the Fog Foundation, the Augustinus Foundation, the Tvergaard Foundation, and the Olga Bryde Foundation.

Duality of Interest. This study was also funded by HypoSafe A/S. A.S. has received research grants from HypoSafe. L.S.R. is an employee at HypoSafe. C.B.J. is a part-time employee at HypoSafe. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. A.S. contributed to the conceptual design, recruited participants, conducted the study, researched the data, performed the data analysis, and wrote the manuscript. T.W.K., B.T., and C.B.J. contributed to the conceptual design and data analysis and wrote the manuscript. U.P.-B. and L.S.R. contributed to the conceptual design and data analysis. S.S.D. and C.S.S.F. contributed in performing the study. L.H., J.F., J.J.H., and M.N.N. contributed to data analysis. L.T. contributed to recruitment and enrollment of participants. All authors contributed significantly to the making of the manuscript, data collecting or analysis, and reviewing and editing of the manuscript. T.W.K. and B.T. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 74th Scientific Sessions of the American Diabetes Association, San Francisco, CA, 13–17 June 2014.